Similarly to the CPU benchmarks, the latency and throughput of a confidential ERC20 token transfer can be measured.
The throughput shown here is the maximum that can be achieved with TFHE-rs on an 8xH100 GPU node, in an ideal scenario where all transactions are independent. In a blockchain protocol, the throughput would be limited by the latency of the network and the necessity to apply other operations (compression, decompression, rerandomization).
